Wolff algorithm: Difference between revisions

Revision as of 13:25, 15 February 2019

Revision as of 04:34, 18 April 2021

Line 1:

The '''Wolff algorithm'''<ref>{{Cite journal|last=Wolff|first=Ulli|date=1989-01-23|title=Collective Monte Carlo Updating for Spin Systems|url=https://link.aps.org/doi/10.1103/PhysRevLett.62.361|journal=Physical Review Letters|volume=62|issue=4|pages=361–364|doi=10.1103/PhysRevLett.62.361}}</ref>, named after [[Ulli Wolff]], is an [[algorithm]] for [[Monte Carlo simulation]] of the [[Ising model]] and [[Potts model]] in which the unit to be flipped is not a single spin, (as in the [[Glauber dynamics|heat bath]] or [[Metropolis–Hastings algorithm|Metropolis algorithms,]]) but a cluster of them. This cluster is defined as the set of neighbouringconnected spins sharing the same value of the spin. The Wolff algorithm is anstates, improvementbased overon the [[Swendsen–WangRandom algorithmcluster model|Fortuin-Kasteleyn representation]]. because it has a larger probability of flipping bigger clusters.

The Wolff algorithm is similar to the [[Swendsen–Wang algorithm]], but different in that the former only flips one randomly chosen cluster with probability 1, while the latter flip every cluster independently with probability 1/2. It is shown numerically that flipping only one cluster decreases the [[autocorrelation]] time of the spin statistics.

The advantage of Wolff algorithm over other algorithms for magnetic spin simulations like single spin flip is that it allows non-local moves on the energy. One important consequence of this is that in some situations (e.g. ferromagnetic Ising model or fully frustrated Ising model), the scaling of the Multicanonic simulation is <math>N^2</math>, better than <math>N^{2+z}</math>, where z is the exponent associated with the critical slowing down phenomena.